Internal validity

Validity is an important consideration in the interpretation of results from epidemiological studies [115]. There are two types of validity: internal and external [116, 117]. Internal validity refers to the strength of the study inferences, which are related to the absence of systematic errors: selection bias, information bias, and confounding [115-117].

Selection bias is present when study participants have different probabilities of being included in the study [115]. For the current study, we selected medical and dental students from two faculties who attended the recruitment lectures. It cannot be argued with certainty that there are systematic differences in relevant study characteristics between the students who did and did not attend the recruitment lectures. The same may be assumed in relation to the medical students who were invited to participate and those from other, smaller faculties and departments of the NSMU who were not invited. For Stage 2, to achieve the desired statistical power, we invited all dental students and a stratified random proportionate sample of medical students (who were a group nearly double in size compared to dental students) from Stage 1. Nonetheless, the final sample was not well balanced, with a lower response rate in medical students (57.6%) than dental students (79.3%) in Stage 2. This may have led to an underestimation of DA and OH problems in medical students. Moreover, the OHIP-14 scores might be positively overestimated due to the overall response rate of 64.9% for Stage 2.

Information bias results from errors in the measurement of study variables [117, 118].

In the present study, data were obtained from the clinical dental examination and from the structured, self-administered questionnaires. The clinical dental examination was performed on all study participants, and information on dental caries experience, oral hygiene, and gingival soft tissue status was recorded. Dental caries experience was measured by the DMFT index, which was documented during the examination according to WHO recommendations [7]. Although the DMFT index has been used for 80 years and

is one of the most common tools used in epidemiological dental studies, it has several limitations [119]. The DMFT index only counts teeth with carious lesions extending into the dentin; enamel carious lesions are not counted, nor is the activity level of carious lesions recorded. Moreover, it was difficult to confirm the reason for tooth extraction at the time of the clinical dental examination. The DMFT index does not count sealants, but can overestimate dental caries experience by taking into consideration teeth with cosmetic restorations. The DMFT calculation gives equal weight to MT, restored teeth, and teeth with untreated dental caries. In addition, in the current study, only visual and tactile methods were applied to detect dental caries; radiographs were not taken, which could lead to an underestimation of dental caries. An Israeli study conducted among participants aged 18-20 years showed that average DMFT index and DT with radiographs were 1.42 and 1.75 higher, respectively, than values obtained without radiographs [120]. Indeed, when radiographs are used, early and secondary proximal dental caries, as well as aesthetic restorations, may be more frequently detected. Nevertheless, radiographic equipment is not always available in many epidemiological studies. Finally, DMFT index may have a skewed distribution in the general population. To solve this problem and focus on individuals with the highest DMFT index, the SiC index can be calculated [102], and that was done in the present study. Oral hygiene was assessed by the OHI-S [103], which has been previously validated and is one of the most commonly used tools in epidemiological studies and clinical practice [103, 121]. The GI was applied to evaluate qualitative changes in the gingival soft tissue [104]. The GI has also gained wide acceptance as a simple, accurate method to assess gingival health in epidemiological and clinical research [122].

When considering the instruments available to measure DA, the DAS and MDAS are the most frequently used tools in young university students. Compared to the DAS, the MDAS has identical response options for all questions (from not anxious to extremely anxious) and includes one additional question about anxiety of dental injection. This item

on injection will probably also reflect general syringe phobia among respondents and blend in with the total score. As the distribution of any kind of phobia is unknown in the young population of North-West Russia, we considered the DAS to be the most appropriate measurement for the present population of medical and dental students. Nevertheless, some researchers maintain that Corah’s DAS does not consider the theoretical structure of DA and that its response categories are not mutually exclusive [30]. In the current study, the Russian version of the DAS seemed to have acceptable psychometric properties. The fact that only three of the 807 DAS respondents omitted one item adds support to the face validity of the instrument, implying that it subjectively appears to measure what it is supposed to measure [123]. Moreover, students who confirmed DA as their reason for not scheduling dental visits had significantly higher DAS scores than students who reported

“other” reasons for not going to a dentist (12.5 vs. 8.5, p<0.001), which provided evidence of criterion validity, i.e., “the degree of correspondence between a test measure and one or more external referents (criteria)” [123].

To assess OHRQoL, we used the Russian language version of the OHIP-14, an instrument that has been validated in another adult Russian population [81]. Although the instrument was validated among middle-aged adults with periodontal diseases, the results of the present study also provide evidence of the good construct validity of OHIP-14 items when applied to young adults; the OHIP-14 scores discriminated significantly between students with good and poor self-assessed OH (mean 3.6 and 6.6, respectively).

Nevertheless, OHRQoL measures, including OHIP-14, have some limitations, as they focus on negative impacts only and define the frequency of impacts of oral diseases, but do not demonstrate their true significance with regard to quality of life [66, 124]. Finally, in the present study, information on OH behaviours, SES, general health, psychological health, and dental aesthetic was self-reported; thus, the possibility of social desirability bias due to under- or over-reporting cannot be ruled out.

When an association between an exposure X and an outcome Y is investigated, we need to assume and check whether there is a third variable (or group of variables) that is associated with both X and Y, and that thus may influence the observed X-Y association.

This third variable is usually designated as a confounding variable (or confounder) [115].

Interaction (or effect modification) exists when the relationship between two variables is different for different levels (or presence/absence) of a third variable [115]. To control for confounders and to assess interactions, multivariate analysis (modelling) and stratification are often used [115]. In all three papers that comprise this thesis, we used multivariable analysis to find adjusted associations between the outcomes of interest (DMFT index, DA, and OHRQoL) and the selected predictors. Moreover, as expected, we found a different level of DA in medical and dental students, and significant interactions between “faculty”

and “mother’s education”, and “faculty” and “regularity of dental visits” in relation to the DAS scores. Given that, we performed the statistical analysis for medical and dental students separately. Nevertheless, the selection of predictors, which should be included in multivariable analyses, is controversial and represents a difficult task in epidemiological analysis [125]. Theoretical or empirical strategies may be used to identify potential confounders or effect modifiers. While theoretical identification is based on results of previous studies or expert knowledge, empirical strategies select factors from the current working dataset [126]. In the present study, we endeavoured to apply both strategies, taking into consideration factors which were found to be significant in other studies, as well as results of univariable analyses, in which the crude associations between outcomes and predictors were determined. Nevertheless, we did not take into account other factors that are potentially associated with the outcomes studied, for example, consumption of sugars including soft drinks, content of fluoride in drinking water, and smoking.